The regulations associated with current on-board diagnostic (OBD) systems demand very strict monitoring of engine performance at certain intervals. This monitoring entails the diagnosis of any fault which could cause the tailpipe emissions of carbon monoxide, unburned hydrocarbons and oxides of nitrogen to rise above legislated values. In the absence of durable and affordable sensors to monitor the concentrations of these gases on the vehicle, coupled with the extremely low emissions limits imposed by future legislation, complying with OBD requirements is an extremely challenging task.
The automotive industry currently uses physical models to predict engine performance in the normal, ‘fault-free’ condition. The outputs from the model are then compared to measured output signals and the difference analysed to determine if a fault has occurred (Fig 1). However, as the emissions thresholds have reduced, the OBD fault detection thresholds have decreased accordingly, thereby increasing the challenge in engine modelling and monitoring. In addition, the expense of physical model–based OBD, in terms of mathematical complexity and computational intensity, has led to research on more practical alternatives.
A multidisciplinary research team at QUB in the Virtual Engineering Centre, drawn from the Intelligent Systems and Control and the Internal Combustion Engines research groups, has pioneered a completely new approach to on-board diagnostic systems for internal combustion engines using multivariate statistical process control. In particular, an Auto-Associative Neural Network (Fig 2) implementation has proved effective in handling the inherent nonlinearities in the data modelling. This approach is capable of analysing the highly dynamic and non-linear signals provided by sensors currently available on production vehicles and it is envisaged that its performance will exceed existing OBD techniques.
For more information please contact
Prof. George Irwin (Intelligent Systems and Control)
Dr Geoff McCullough (Internal Combustion Engines)
Prof GW Irwin, ‘Automotive engine fault detection’, Invited lecture at EPSRC Winter School on Data Modelling, University of Sheffield, 21-25 January 2008.
Dr N McDowell, ‘Fault diagnosis for internal combustion engines - current and future techniques’, Invited lecture at 3rd IET Conference on Automotive Electronics, University of Warwick, Coventry, 28-29 June 2007.
Prof GW Irwin and Dr U Kruger,‘Intelligent use of data for condition monitoring and applications’, Invited Plenary, 2nd International Conference on Intelligent on Intelligent Computing (ICIC ’06), Kunming, Yunnan Province, China, August 2006.
Dr U Kruger, ‘Development and Application of nonlinear PCA for fault diagnosis in internal combustion engines’, Invited Speaker, Workshop on ‘Principal Manifolds for Data Cartography and Data Reduction’, Department of Mathematics, University of Leicester, 24-26 August 2006.
Fig 3. Instrumented 4-cyclinder, 1.8 litre Nissan petrol engine and transient dynamometer for conducting engine tests
Fig 4. Air leak fault in engine manifold and locations of engine sensors
The researchers gratefully acknowledge the Engineering and Physical Sciences Research Council (EPSRC) for funding this project (EP/C005457/1).